Climate Impacts on Agriculture Production
Nicholas A. Potter
Oct 03, 2017
Acknowledgements
- Michael Brady
- Kirti Rajagopalan
- Jennifer Adams
- Chad Kruger (CSANR)
Estimating climate impacts
An excellent review is Blanc and Reilly (2017). There are three broad approaches:
- Cross-sectional estimation of production effect of temperature (Mendelsohn, Nordhaus, and Shaw 1994).
- Estimate production using panel data with temperature as exogenous variable (Schlenker and Roberts 2009).
- Crop simulation models of how crops respond to temperature changes (Antle and Stöckle 2017).
Mendelsohn, Nordhaus, and Shaw (1994)
Schlenker and Roberts (2009)
Data
- 4km grid of t-max, t-min, and precip. Historical: 1979-2015, Projected: 2006-2099 (Jenny Adam’s Lab).
- County level: ag land rent, crop and pasture acres, irrigated acres, income, population (NASS, census)
Naive matching
Treatment effect: \(E[Y|X, T = 1] - E[Y|X, T = 0]\). But this is not an RCT. Matching provides a method of balancing \(X\) between treatment and control groups. If \(c(i)\) is the control that minimizes \(d(X_i, X_{c(i)})\) for some distance metric \(d\), then a pair matching estimator is
\[\hat{\tau}_{1p} = \frac{1}{N_1} \sum \limits_{i \in N_1} (Y_{i,1} - Y_{c(i), 0}),\]
where \(Y_{i,1}\) is the treated observation and \(Y_{c(i),0}\) is the matched control observation.
There be dragons
Problem: But here \(Y_{i,1}\) is unobserved while \(Y_{i,0}\) is observed…
Solution: Estimate \(Y_{i,1}\) by matching the treatment characteristics \(T_i\) to the closest treatment characteristics in the control, i.e. choose \(c(i) = \mbox{argmin}_{c(i)} \; d(T_i, T_{c(i)})\), in which case the estimator is
\[\hat{\tau}_{1p} = \frac{1}{N_1} \sum \limits_{i \in N_1} (Y_{c(i),0} - Y_{i, 0}).\]
\(T_i\) here is not a dummy. It’s a vector of continuous climate variables. Other predictors of \(Y_{i,1}\) such as random forests may be better…
Distance metrics
Propensity score matching: good at minimizing distance along a single dimension that is a combination of all other dimensions.
Mahalanobis distance matching: good at minimizing the sum of distances between individual coordinates of \(X\).
Genetic matching: search space over a weighted combination of the two. Enter
RpackageMatching(Sekhon 2011).
Issue 1: Multiple causal mechanisms
Issue 2: How good are the matches?
2.a Are the climate variables \(T_i\) actually good predictors of ag land rent value and other outcomes? Potential check: use historical climate and outcomes \((Y_{i, -1} | T_{i,-1})\) to predict current outcomes \((Y_{i, 0} | T_{i,0})\) and check accuracy.
2.b Does matched climate look like treatment climate? Potential checks: look at balance of matched variables, map of difference in climate measures between treated and matched controls, etc…
References
Antle, John M., and Claudio O. Stöckle. 2017. “Climate Impacts on Agriculture: Insights from Agronomic-Economic Analysis.” Review of Environmental Economics and Policy 11 (2). Oxford University Press (OUP): 299–318. doi:10.1093/reep/rex012.
Blanc, Elodie, and John Reilly. 2017. “Approaches to Assessing Climate Change Impacts on Agriculture: An Overview of the Debate.” Review of Environmental Economics and Policy 11 (2). Oxford University Press (OUP): 247–57. doi:10.1093/reep/rex011.
Mendelsohn, Robert, William D Nordhaus, and Daigee Shaw. 1994. “The Impact of Global Warming on Agriculture: A Ricardian Analysis.” The American Economic Review. JSTOR, 753–71.
Schlenker, W., and M. J. Roberts. 2009. “Nonlinear Temperature Effects Indicate Severe Damages to U.s. Crop Yields Under Climate Change.” Proceedings of the National Academy of Sciences 106 (37). Proceedings of the National Academy of Sciences: 15594–8. doi:10.1073/pnas.0906865106.
Sekhon, Jasjeet S. 2011. “Multivariate and Propensity Score Matching Software with Automated Balance Optimization: The Matching Package for R.” Journal of Statistical Software 42 (7): 1–52.